Ramp's latest breakdown of enterprise AI spending shows Fable 5's share of corporate Anthropic budgets stuck at roughly 11% for several consecutive months, while Opus 5 has climbed past it to claim a larger slice of the same wallet. According to Techmeme, the shift is part of a broader pattern in Ramp's data: companies are steering spend toward the models that cost less to run at scale, not necessarily the ones topping leaderboard charts.

Ramp tracks this because it processes corporate card transactions and vendor invoices for tens of thousands of businesses, which gives it a real-time view of what companies are actually paying for AI tools — as opposed to what they say they're evaluating in a survey. That distinction matters: a model can win headlines on launch day and still lose wallet share within a quarter if it doesn't hold up on cost per task.

For teams building on Anthropic's stack, the plateau is the more interesting number. A stall at ~11% doesn't mean Fable 5 usage dropped — it means it stopped growing relative to everything else companies are buying, at exactly the moment Opus 5 started eating into that same budget line.

What the spend data is actually measuring

Ramp's index is not a satisfaction score or a capability benchmark. It's a ledger: which model line shows up on invoices, how much of it, and whether that share is rising or falling month over month. That makes it one of the few data sources that reflects renewal decisions rather than launch-week enthusiasm.

A few things typically drive a plateau like Fable 5's:

None of that requires Fable 5 to be a worse model. It only requires that, at the margin, another option offers better economics for the tasks companies are actually running in production.

Why this matters more than another benchmark

Benchmarks measure what a model can do under controlled conditions. Corporate card data measures what finance and engineering teams were willing to keep paying for after the first invoice arrived. The two rankings don't have to agree, and increasingly they don't.

Anthropic itself sells across multiple price tiers for a reason — different tasks warrant different cost structures. What Ramp's numbers suggest is that buyers are getting more disciplined about matching the tier to the task, rather than defaulting to whatever model launched most recently or generated the most buzz.

What builders should take from this

If you're choosing between Anthropic models for a production workload, the practical questions this data raises are:

Teams that treat model selection as a one-time decision are the ones most likely to end up on the losing side of a plateau like Fable 5's — paying for a tier they chose under different price and capability conditions than the ones they're operating in now.

AiiN's takeaway

Spend data is a lagging indicator, but it's a harder one to fake than a benchmark result. Companies don't keep paying for something that isn't earning its cost, and they don't switch away from something that is. When a widely-adopted model's share of the budget stalls while a rival's climbs, that's the market doing quiet due diligence that no leaderboard captures.

Our take: expect procurement teams to keep pushing this trend further. As enterprise AI budgets mature past the pilot phase, cost-per-outcome is becoming the metric that actually decides renewals — and, in our estimation, that pressure will keep favoring whichever model line offers the best economics at a given capability level, regardless of which one shipped most recently.